ipectral differences 
• Saturation increase 
tufmann 1985). 
lputing and stretch 
ssion via photowrite 
i offers optimum 
structural mapping 
; optimization pro- 
:ed image product, 
I procedure the 
evaluation results 
e computer. Most of 
re working pixel- 
formation. The un 
pixels with similar 
ar example with the 
re space as a cri- 
The individual 
class whose center 
smaller than a 
vised methods accu- 
calculate the 
ses before classi- 
II pixels to one 
be carried out using 
The well known 
example, is based 
ihood with which 
rs of these classes, 
ed to the class with 
AVIS, 1978). To get 
criptions of the 
n values in each 
ix, are needed, 
tistics training 
ntroduced. The main 
dures is the se- 
e sample areas for 
ds of classification 
;upervised pro- 
ir the classi- 
escribe the 
,s. A multispectral 
latures for example, 
;tural features as 
le textural parame- 
?ach pixel in a 
.. Well known textu- 
itistical evaluation 
?n neighouring pixels 
, 1973). The calcula- 
jmeters for each of 
jds to 7 additional 
luent supervised 
ion, with both the 
jres, a highly so- 
er channel combina- 
out. Furthermore, 
arocedure (WU, 1975) 
jll information con- 
jral channels in a 
lassification of the 
euclidean distance 
ing channels 3/4. 
ce (d=20 grey values) 
11 be discriminated, 
more spectral classes 
tic Mapper data 
te 7.7.1984) a small 
ms to be a favourable 
. Fig. 4 demonstrates 
eads to more than 10